3D Object Detection in LIDAR Point Cloud Based on Background Subtraction

Abstract

Autonomous vehicles have a key role in transportation systems of the future, but there are still many difficulties to overcome. Nowadays one of the most critical problems in autonomous driving is the precise and robust detection of traffic participants. This paper presents a LIDAR-based 3D object detection method. The algorithm uses HD Map to subtract the static background points from the LIDAR point cloud. The remaining points are grouped by clustering, then 3D boxes are fitted to the clusters. The object detection method presented in this paper was tested on real sensor data collected by a solid-state LIDAR on the highway module of the ZalaZONE proving ground. The results showed that the developed algorithm performs as intended in a highway scenario, detecting vehicles even more than 100 meters away from the sensor by a framerate of ~20FPS.